Proceedings of the Annual Meeting of the Cognitive Science Society

It is inconceivable how chaotic the world would look to hu-mans, faced with innumerable decisions a day to be made un-der uncertainty, had they been lacking the capacity to distin-guish the relevant from the irrelevant—a capacity which com-putationally amounts to handling probabilistic independencerelations. The highly parallel and distributed computationalmachinery of the brain suggests that a satisfying process-levelaccount of human independence judgment should also mimicthese features. In this work, we present the first rational, dis-tributed, message-passing, process-level account of indepen-dence judgment, called D∗. Interestingly, D∗ shows a curi-ous, but normatively justified tendency for quick detection ofdependencies, whenever they hold. Furthermore, D∗ outper-forms all the previously proposed algorithms in the AI litera-ture in terms of worst-case running time, and a salient aspectof it is supported by recent work in neuroscience investigatingpossible implementations of Bayes nets at the neural level. D∗exemplifies how the pursuit of cognitive plausibility can leadto the discovery of state-of-the-art algorithms with appealingproperties, and its simplicity makes D∗ potentially a good can-didate as a teaching tool.